Converting RGB into Grayscale

[brunoj]

Hello. I have 3 folders with images subfolders as train, test and validate. I have used T.Grayscale(1) to convert into grayscale but when I show the image , its like this. I dont understand why? Here is the image which Im getting

my code is here:
import matplotlib.pyplot as plt
import torch.nn.functional as F
import torch
import numpy as np

def show_image(image,label):

image = image.permute(1,2,0)
plt.imshow(image)
plt.title(label)

def show_grid(image,title = None):

image = image.permute(1,2,0)


plt.figure(figsize=[15, 15])
plt.imshow(image)
if title != None:
    plt.title(title)

def accuracy(y_pred,y_true):
y_pred = F.softmax(y_pred,dim = 1)
top_p,top_class = y_pred.topk(1,dim = 1)
equals = top_class == y_true.view(*top_class.shape)
return torch.mean(equals.type(torch.FloatTensor))

def view_classify(image,ps,label):

class_name = ['F', 'M' ,'N', 'S', 'Q','V']
classes = np.array(class_name)

ps = ps.cpu().data.numpy().squeeze()

image = image.permute(1,2,0)

fig, (ax1, ax2) = plt.subplots(figsize=(8,12), ncols=2)
ax1.imshow(img)
ax1.set_title('Ground Truth : {}'.format(class_name[label]))
ax1.axis('off')
ax2.barh(classes, ps)
ax2.set_aspect(0.1)
ax2.set_yticks(classes)
ax2.set_yticklabels(classes)
ax2.set_title('Predicted Class')
ax2.set_xlim(0, 1.1)

plt.tight_layout()

return None

train_transform = T.Compose([

                         T.Resize(size=(CFG.img_size,CFG.img_size)), 
                         T.ToTensor(), 
                         T.Grayscale(num_output_channels=1)

])

validate_transform = T.Compose([

                         T.Resize(size=(CFG.img_size,CFG.img_size)), 
                        
                         T.ToTensor(),
                         T.Grayscale(num_output_channels=1)

])

test_transform = T.Compose([

                         T.Resize(size=(CFG.img_size,CFG.img_size)), # Resizing the image to be 32 by 22
                         
                         T.ToTensor(), 
                         T.Grayscale(num_output_channels=1)

])

[brunoj]

The original images are like thid. But above output imge gets blurry as well. I understand i changed the size like 32x32 from original size 438x288 so maybe thats the reason it is blurry.

[ptrblck]

matplotlib.pyplot.plot uses the viridis colormap by default so you would need to use cmap="gray" to show the image in grayscale:

img = np.random.randint(0, 256, (32, 32)).astype(np.uint8)
plt.imshow(img)

plt.imshow(img, cmap="gray")

Yes, resizing the image to 32x32 pixels will create these artifacts.

[brunoj]

Thankyou so much. This worked for me. I have trained my model, saved the weights and tested now Im trying to use view_classify from a tutorial but it is giving me errors like invalid shape for image data. What can I do? I searched it here people said copy the helper.py and add i dont know where is helper.py as I just used the tutorial but nothing is mentioned there. Someone on the forum pasted this code:
import matplotlib.pyplot as plt
import numpy as np

def view_classify(img, ps, version=“MNIST”):
‘’’ Function for viewing an image and it’s predicted classes.
‘’’
ps = ps.data.numpy().squeeze()

fig, (ax1, ax2) = plt.subplots(figsize=(6,9), ncols=2)
ax1.imshow(img.resize_(1, 28, 28).numpy().squeeze())
ax1.axis('off')
ax2.barh(np.arange(10), ps)
ax2.set_aspect(0.1)
ax2.set_yticks(np.arange(10))
if version == "MNIST":
    ax2.set_yticklabels(np.arange(10))
elif version == "Fashion":
    ax2.set_yticklabels(['T-shirt/top',
                        'Trouser',
                        'Pullover',
                        'Dress',
                        'Coat',
                        'Sandal',
                        'Shirt',
                        'Sneaker',
                        'Bag',
                        'Ankle Boot'], size='small');
ax2.set_title('Class Probability')
ax2.set_xlim(0, 1.1)

plt.tight_layout()

I saved it as view_classify.ipynb and uploaded to colab. like this:

And I got error in the image above. Plz tell how to fix asap. My dataset is from kaggle ecg data images and 32x32x1 CHANNEL.

[brunoj]

Train an EfficientNet Model in PyTorch for Medical Diagnosis | by Yaokun Lin @ MachineLearningQuickNotes | Geek Culture | Medium This is the tutorial that Im following.

[ptrblck]

I would guess channels-last data layout is expected so permute your tensor into [height, width, channels] and it should work.